1 // Copyright 2009 The Go Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style
3 // license that can be found in the LICENSE file.
5 // Garbage collector: sweeping
7 // The sweeper consists of two different algorithms:
9 // * The object reclaimer finds and frees unmarked slots in spans. It
10 // can free a whole span if none of the objects are marked, but that
11 // isn't its goal. This can be driven either synchronously by
12 // mcentral.cacheSpan for mcentral spans, or asynchronously by
13 // sweepone, which looks at all the mcentral lists.
15 // * The span reclaimer looks for spans that contain no marked objects
16 // and frees whole spans. This is a separate algorithm because
17 // freeing whole spans is the hardest task for the object reclaimer,
18 // but is critical when allocating new spans. The entry point for
19 // this is mheap_.reclaim and it's driven by a sequential scan of
20 // the page marks bitmap in the heap arenas.
22 // Both algorithms ultimately call mspan.sweep, which sweeps a single
28 "runtime/internal/atomic"
34 // State of background sweep.
35 type sweepdata struct {
44 // centralIndex is the current unswept span class.
45 // It represents an index into the mcentral span
46 // sets. Accessed and updated via its load and
47 // update methods. Not protected by a lock.
49 // Reset at mark termination.
50 // Used by mheap.nextSpanForSweep.
51 centralIndex sweepClass
54 // sweepClass is a spanClass and one bit to represent whether we're currently
55 // sweeping partial or full spans.
56 type sweepClass uint32
59 numSweepClasses = numSpanClasses * 2
60 sweepClassDone sweepClass = sweepClass(^uint32(0))
63 func (s *sweepClass) load() sweepClass {
64 return sweepClass(atomic.Load((*uint32)(s)))
67 func (s *sweepClass) update(sNew sweepClass) {
68 // Only update *s if its current value is less than sNew,
69 // since *s increases monotonically.
71 for sOld < sNew && !atomic.Cas((*uint32)(s), uint32(sOld), uint32(sNew)) {
74 // TODO(mknyszek): This isn't the only place we have
75 // an atomic monotonically increasing counter. It would
76 // be nice to have an "atomic max" which is just implemented
77 // as the above on most architectures. Some architectures
78 // like RISC-V however have native support for an atomic max.
81 func (s *sweepClass) clear() {
82 atomic.Store((*uint32)(s), 0)
85 // split returns the underlying span class as well as
86 // whether we're interested in the full or partial
87 // unswept lists for that class, indicated as a boolean
88 // (true means "full").
89 func (s sweepClass) split() (spc spanClass, full bool) {
90 return spanClass(s >> 1), s&1 == 0
93 // nextSpanForSweep finds and pops the next span for sweeping from the
94 // central sweep buffers. It returns ownership of the span to the caller.
95 // Returns nil if no such span exists.
96 func (h *mheap) nextSpanForSweep() *mspan {
98 for sc := sweep.centralIndex.load(); sc < numSweepClasses; sc++ {
99 spc, full := sc.split()
100 c := &h.central[spc].mcentral
103 s = c.fullUnswept(sg).pop()
105 s = c.partialUnswept(sg).pop()
108 // Write down that we found something so future sweepers
109 // can start from here.
110 sweep.centralIndex.update(sc)
114 // Write down that we found nothing.
115 sweep.centralIndex.update(sweepClassDone)
119 // finishsweep_m ensures that all spans are swept.
121 // The world must be stopped. This ensures there are no sweeps in
125 func finishsweep_m() {
126 // Sweeping must be complete before marking commences, so
127 // sweep any unswept spans. If this is a concurrent GC, there
128 // shouldn't be any spans left to sweep, so this should finish
129 // instantly. If GC was forced before the concurrent sweep
130 // finished, there may be spans to sweep.
131 for sweepone() != ^uintptr(0) {
135 if go115NewMCentralImpl {
136 // Reset all the unswept buffers, which should be empty.
137 // Do this in sweep termination as opposed to mark termination
138 // so that we can catch unswept spans and reclaim blocks as
140 sg := mheap_.sweepgen
141 for i := range mheap_.central {
142 c := &mheap_.central[i].mcentral
143 c.partialUnswept(sg).reset()
144 c.fullUnswept(sg).reset()
148 // Sweeping is done, so if the scavenger isn't already awake,
149 // wake it up. There's definitely work for it to do at this
153 nextMarkBitArenaEpoch()
156 func bgsweep(c chan int) {
159 lockInit(&sweep.lock, lockRankSweep)
163 goparkunlock(&sweep.lock, waitReasonGCSweepWait, traceEvGoBlock, 1)
166 for sweepone() != ^uintptr(0) {
170 for freeSomeWbufs(true) {
175 // This can happen if a GC runs between
176 // gosweepone returning ^0 above
177 // and the lock being acquired.
182 goparkunlock(&sweep.lock, waitReasonGCSweepWait, traceEvGoBlock, 1)
186 // sweepone sweeps some unswept heap span and returns the number of pages returned
187 // to the heap, or ^uintptr(0) if there was nothing to sweep.
188 func sweepone() uintptr {
190 sweepRatio := mheap_.sweepPagesPerByte // For debugging
192 // increment locks to ensure that the goroutine is not preempted
193 // in the middle of sweep thus leaving the span in an inconsistent state for next GC
195 if atomic.Load(&mheap_.sweepdone) != 0 {
199 atomic.Xadd(&mheap_.sweepers, +1)
201 // Find a span to sweep.
203 sg := mheap_.sweepgen
205 if go115NewMCentralImpl {
206 s = mheap_.nextSpanForSweep()
208 s = mheap_.sweepSpans[1-sg/2%2].pop()
211 atomic.Store(&mheap_.sweepdone, 1)
214 if state := s.state.get(); state != mSpanInUse {
215 // This can happen if direct sweeping already
216 // swept this span, but in that case the sweep
217 // generation should always be up-to-date.
218 if !(s.sweepgen == sg || s.sweepgen == sg+3) {
219 print("runtime: bad span s.state=", state, " s.sweepgen=", s.sweepgen, " sweepgen=", sg, "\n")
220 throw("non in-use span in unswept list")
224 if s.sweepgen == sg-2 && atomic.Cas(&s.sweepgen, sg-2, sg-1) {
229 // Sweep the span we found.
230 npages := ^uintptr(0)
234 // Whole span was freed. Count it toward the
235 // page reclaimer credit since these pages can
236 // now be used for span allocation.
237 atomic.Xadduintptr(&mheap_.reclaimCredit, npages)
239 // Span is still in-use, so this returned no
240 // pages to the heap and the span needs to
241 // move to the swept in-use list.
246 // Decrement the number of active sweepers and if this is the
247 // last one print trace information.
248 if atomic.Xadd(&mheap_.sweepers, -1) == 0 && atomic.Load(&mheap_.sweepdone) != 0 {
249 // Since the sweeper is done, move the scavenge gen forward (signalling
250 // that there's new work to do) and wake the scavenger.
252 // The scavenger is signaled by the last sweeper because once
253 // sweeping is done, we will definitely have useful work for
254 // the scavenger to do, since the scavenger only runs over the
255 // heap once per GC cyle. This update is not done during sweep
256 // termination because in some cases there may be a long delay
257 // between sweep done and sweep termination (e.g. not enough
258 // allocations to trigger a GC) which would be nice to fill in
259 // with scavenging work.
262 mheap_.pages.scavengeStartGen()
265 // Since we might sweep in an allocation path, it's not possible
266 // for us to wake the scavenger directly via wakeScavenger, since
267 // it could allocate. Ask sysmon to do it for us instead.
270 if debug.gcpacertrace > 0 {
271 print("pacer: sweep done at heap size ", memstats.heap_live>>20, "MB; allocated ", (memstats.heap_live-mheap_.sweepHeapLiveBasis)>>20, "MB during sweep; swept ", mheap_.pagesSwept, " pages at ", sweepRatio, " pages/byte\n")
278 // isSweepDone reports whether all spans are swept or currently being swept.
280 // Note that this condition may transition from false to true at any
281 // time as the sweeper runs. It may transition from true to false if a
282 // GC runs; to prevent that the caller must be non-preemptible or must
283 // somehow block GC progress.
284 func isSweepDone() bool {
285 return mheap_.sweepdone != 0
288 // Returns only when span s has been swept.
290 func (s *mspan) ensureSwept() {
291 // Caller must disable preemption.
292 // Otherwise when this function returns the span can become unswept again
293 // (if GC is triggered on another goroutine).
295 if _g_.m.locks == 0 && _g_.m.mallocing == 0 && _g_ != _g_.m.g0 {
296 throw("mspan.ensureSwept: m is not locked")
299 sg := mheap_.sweepgen
300 spangen := atomic.Load(&s.sweepgen)
301 if spangen == sg || spangen == sg+3 {
304 // The caller must be sure that the span is a mSpanInUse span.
305 if atomic.Cas(&s.sweepgen, sg-2, sg-1) {
309 // unfortunate condition, and we don't have efficient means to wait
311 spangen := atomic.Load(&s.sweepgen)
312 if spangen == sg || spangen == sg+3 {
319 // Sweep frees or collects finalizers for blocks not marked in the mark phase.
320 // It clears the mark bits in preparation for the next GC round.
321 // Returns true if the span was returned to heap.
322 // If preserve=true, don't return it to heap nor relink in mcentral lists;
323 // caller takes care of it.
324 func (s *mspan) sweep(preserve bool) bool {
325 if !go115NewMCentralImpl {
326 return s.oldSweep(preserve)
328 // It's critical that we enter this function with preemption disabled,
329 // GC must not start while we are in the middle of this function.
331 if _g_.m.locks == 0 && _g_.m.mallocing == 0 && _g_ != _g_.m.g0 {
332 throw("mspan.sweep: m is not locked")
334 sweepgen := mheap_.sweepgen
335 if state := s.state.get(); state != mSpanInUse || s.sweepgen != sweepgen-1 {
336 print("mspan.sweep: state=", state, " sweepgen=", s.sweepgen, " mheap.sweepgen=", sweepgen, "\n")
337 throw("mspan.sweep: bad span state")
341 traceGCSweepSpan(s.npages * _PageSize)
344 atomic.Xadd64(&mheap_.pagesSwept, int64(s.npages))
349 c := _g_.m.p.ptr().mcache
351 // The allocBits indicate which unmarked objects don't need to be
352 // processed since they were free at the end of the last GC cycle
353 // and were not allocated since then.
354 // If the allocBits index is >= s.freeindex and the bit
355 // is not marked then the object remains unallocated
356 // since the last GC.
357 // This situation is analogous to being on a freelist.
359 // Unlink & free special records for any objects we're about to free.
360 // Two complications here:
361 // 1. An object can have both finalizer and profile special records.
362 // In such case we need to queue finalizer for execution,
363 // mark the object as live and preserve the profile special.
364 // 2. A tiny object can have several finalizers setup for different offsets.
365 // If such object is not marked, we need to queue all finalizers at once.
366 // Both 1 and 2 are possible at the same time.
367 hadSpecials := s.specials != nil
368 specialp := &s.specials
371 // A finalizer can be set for an inner byte of an object, find object beginning.
372 objIndex := uintptr(special.offset) / size
373 p := s.base() + objIndex*size
374 mbits := s.markBitsForIndex(objIndex)
375 if !mbits.isMarked() {
376 // This object is not marked and has at least one special record.
377 // Pass 1: see if it has at least one finalizer.
379 endOffset := p - s.base() + size
380 for tmp := special; tmp != nil && uintptr(tmp.offset) < endOffset; tmp = tmp.next {
381 if tmp.kind == _KindSpecialFinalizer {
382 // Stop freeing of object if it has a finalizer.
383 mbits.setMarkedNonAtomic()
388 // Pass 2: queue all finalizers _or_ handle profile record.
389 for special != nil && uintptr(special.offset) < endOffset {
390 // Find the exact byte for which the special was setup
391 // (as opposed to object beginning).
392 p := s.base() + uintptr(special.offset)
393 if special.kind == _KindSpecialFinalizer || !hasFin {
394 // Splice out special record.
396 special = special.next
398 freespecial(y, unsafe.Pointer(p), size)
400 // This is profile record, but the object has finalizers (so kept alive).
401 // Keep special record.
402 specialp = &special.next
407 // object is still live: keep special record
408 specialp = &special.next
412 if hadSpecials && s.specials == nil {
416 if debug.allocfreetrace != 0 || debug.clobberfree != 0 || raceenabled || msanenabled {
417 // Find all newly freed objects. This doesn't have to
418 // efficient; allocfreetrace has massive overhead.
419 mbits := s.markBitsForBase()
420 abits := s.allocBitsForIndex(0)
421 for i := uintptr(0); i < s.nelems; i++ {
422 if !mbits.isMarked() && (abits.index < s.freeindex || abits.isMarked()) {
423 x := s.base() + i*s.elemsize
424 if debug.allocfreetrace != 0 {
425 tracefree(unsafe.Pointer(x), size)
427 if debug.clobberfree != 0 {
428 clobberfree(unsafe.Pointer(x), size)
431 racefree(unsafe.Pointer(x), size)
434 msanfree(unsafe.Pointer(x), size)
442 // Check for zombie objects.
443 if s.freeindex < s.nelems {
444 // Everything < freeindex is allocated and hence
445 // cannot be zombies.
447 // Check the first bitmap byte, where we have to be
448 // careful with freeindex.
450 if (*s.gcmarkBits.bytep(obj / 8)&^*s.allocBits.bytep(obj / 8))>>(obj%8) != 0 {
453 // Check remaining bytes.
454 for i := obj/8 + 1; i < divRoundUp(s.nelems, 8); i++ {
455 if *s.gcmarkBits.bytep(i)&^*s.allocBits.bytep(i) != 0 {
461 // Count the number of free objects in this span.
462 nalloc := uint16(s.countAlloc())
463 nfreed := s.allocCount - nalloc
464 if nalloc > s.allocCount {
465 // The zombie check above should have caught this in
467 print("runtime: nelems=", s.nelems, " nalloc=", nalloc, " previous allocCount=", s.allocCount, " nfreed=", nfreed, "\n")
468 throw("sweep increased allocation count")
471 s.allocCount = nalloc
472 s.freeindex = 0 // reset allocation index to start of span.
474 getg().m.p.ptr().traceReclaimed += uintptr(nfreed) * s.elemsize
477 // gcmarkBits becomes the allocBits.
478 // get a fresh cleared gcmarkBits in preparation for next GC
479 s.allocBits = s.gcmarkBits
480 s.gcmarkBits = newMarkBits(s.nelems)
482 // Initialize alloc bits cache.
483 s.refillAllocCache(0)
485 // The span must be in our exclusive ownership until we update sweepgen,
486 // check for potential races.
487 if state := s.state.get(); state != mSpanInUse || s.sweepgen != sweepgen-1 {
488 print("mspan.sweep: state=", state, " sweepgen=", s.sweepgen, " mheap.sweepgen=", sweepgen, "\n")
489 throw("mspan.sweep: bad span state after sweep")
491 if s.sweepgen == sweepgen+1 || s.sweepgen == sweepgen+3 {
492 throw("swept cached span")
495 // We need to set s.sweepgen = h.sweepgen only when all blocks are swept,
496 // because of the potential for a concurrent free/SetFinalizer.
498 // But we need to set it before we make the span available for allocation
499 // (return it to heap or mcentral), because allocation code assumes that a
500 // span is already swept if available for allocation.
502 // Serialization point.
503 // At this point the mark bits are cleared and allocation ready
504 // to go so release the span.
505 atomic.Store(&s.sweepgen, sweepgen)
507 if spc.sizeclass() != 0 {
508 // Handle spans for small objects.
510 // Only mark the span as needing zeroing if we've freed any
511 // objects, because a fresh span that had been allocated into,
512 // wasn't totally filled, but then swept, still has all of its
513 // free slots zeroed.
515 c.local_nsmallfree[spc.sizeclass()] += uintptr(nfreed)
518 // The caller may not have removed this span from whatever
519 // unswept set its on but taken ownership of the span for
520 // sweeping by updating sweepgen. If this span still is in
521 // an unswept set, then the mcentral will pop it off the
522 // set, check its sweepgen, and ignore it.
524 // Free totally free span directly back to the heap.
528 // Return span back to the right mcentral list.
529 if uintptr(nalloc) == s.nelems {
530 mheap_.central[spc].mcentral.fullSwept(sweepgen).push(s)
532 mheap_.central[spc].mcentral.partialSwept(sweepgen).push(s)
535 } else if !preserve {
536 // Handle spans for large objects.
538 // Free large object span to heap.
540 // NOTE(rsc,dvyukov): The original implementation of efence
541 // in CL 22060046 used sysFree instead of sysFault, so that
542 // the operating system would eventually give the memory
543 // back to us again, so that an efence program could run
544 // longer without running out of memory. Unfortunately,
545 // calling sysFree here without any kind of adjustment of the
546 // heap data structures means that when the memory does
547 // come back to us, we have the wrong metadata for it, either in
548 // the mspan structures or in the garbage collection bitmap.
549 // Using sysFault here means that the program will run out of
550 // memory fairly quickly in efence mode, but at least it won't
551 // have mysterious crashes due to confused memory reuse.
552 // It should be possible to switch back to sysFree if we also
553 // implement and then call some kind of mheap.deleteSpan.
554 if debug.efence > 0 {
555 s.limit = 0 // prevent mlookup from finding this span
556 sysFault(unsafe.Pointer(s.base()), size)
561 c.local_largefree += size
565 // Add a large span directly onto the full+swept list.
566 mheap_.central[spc].mcentral.fullSwept(sweepgen).push(s)
571 // Sweep frees or collects finalizers for blocks not marked in the mark phase.
572 // It clears the mark bits in preparation for the next GC round.
573 // Returns true if the span was returned to heap.
574 // If preserve=true, don't return it to heap nor relink in mcentral lists;
575 // caller takes care of it.
577 // For !go115NewMCentralImpl.
578 func (s *mspan) oldSweep(preserve bool) bool {
579 // It's critical that we enter this function with preemption disabled,
580 // GC must not start while we are in the middle of this function.
582 if _g_.m.locks == 0 && _g_.m.mallocing == 0 && _g_ != _g_.m.g0 {
583 throw("mspan.sweep: m is not locked")
585 sweepgen := mheap_.sweepgen
586 if state := s.state.get(); state != mSpanInUse || s.sweepgen != sweepgen-1 {
587 print("mspan.sweep: state=", state, " sweepgen=", s.sweepgen, " mheap.sweepgen=", sweepgen, "\n")
588 throw("mspan.sweep: bad span state")
592 traceGCSweepSpan(s.npages * _PageSize)
595 atomic.Xadd64(&mheap_.pagesSwept, int64(s.npages))
601 c := _g_.m.p.ptr().mcache
604 // The allocBits indicate which unmarked objects don't need to be
605 // processed since they were free at the end of the last GC cycle
606 // and were not allocated since then.
607 // If the allocBits index is >= s.freeindex and the bit
608 // is not marked then the object remains unallocated
609 // since the last GC.
610 // This situation is analogous to being on a freelist.
612 // Unlink & free special records for any objects we're about to free.
613 // Two complications here:
614 // 1. An object can have both finalizer and profile special records.
615 // In such case we need to queue finalizer for execution,
616 // mark the object as live and preserve the profile special.
617 // 2. A tiny object can have several finalizers setup for different offsets.
618 // If such object is not marked, we need to queue all finalizers at once.
619 // Both 1 and 2 are possible at the same time.
620 hadSpecials := s.specials != nil
621 specialp := &s.specials
624 // A finalizer can be set for an inner byte of an object, find object beginning.
625 objIndex := uintptr(special.offset) / size
626 p := s.base() + objIndex*size
627 mbits := s.markBitsForIndex(objIndex)
628 if !mbits.isMarked() {
629 // This object is not marked and has at least one special record.
630 // Pass 1: see if it has at least one finalizer.
632 endOffset := p - s.base() + size
633 for tmp := special; tmp != nil && uintptr(tmp.offset) < endOffset; tmp = tmp.next {
634 if tmp.kind == _KindSpecialFinalizer {
635 // Stop freeing of object if it has a finalizer.
636 mbits.setMarkedNonAtomic()
641 // Pass 2: queue all finalizers _or_ handle profile record.
642 for special != nil && uintptr(special.offset) < endOffset {
643 // Find the exact byte for which the special was setup
644 // (as opposed to object beginning).
645 p := s.base() + uintptr(special.offset)
646 if special.kind == _KindSpecialFinalizer || !hasFin {
647 // Splice out special record.
649 special = special.next
651 freespecial(y, unsafe.Pointer(p), size)
653 // This is profile record, but the object has finalizers (so kept alive).
654 // Keep special record.
655 specialp = &special.next
660 // object is still live: keep special record
661 specialp = &special.next
665 if go115NewMarkrootSpans && hadSpecials && s.specials == nil {
669 if debug.allocfreetrace != 0 || debug.clobberfree != 0 || raceenabled || msanenabled {
670 // Find all newly freed objects. This doesn't have to
671 // efficient; allocfreetrace has massive overhead.
672 mbits := s.markBitsForBase()
673 abits := s.allocBitsForIndex(0)
674 for i := uintptr(0); i < s.nelems; i++ {
675 if !mbits.isMarked() && (abits.index < s.freeindex || abits.isMarked()) {
676 x := s.base() + i*s.elemsize
677 if debug.allocfreetrace != 0 {
678 tracefree(unsafe.Pointer(x), size)
680 if debug.clobberfree != 0 {
681 clobberfree(unsafe.Pointer(x), size)
684 racefree(unsafe.Pointer(x), size)
687 msanfree(unsafe.Pointer(x), size)
695 // Count the number of free objects in this span.
696 nalloc := uint16(s.countAlloc())
697 if spc.sizeclass() == 0 && nalloc == 0 {
701 nfreed := s.allocCount - nalloc
702 if nalloc > s.allocCount {
703 print("runtime: nelems=", s.nelems, " nalloc=", nalloc, " previous allocCount=", s.allocCount, " nfreed=", nfreed, "\n")
704 throw("sweep increased allocation count")
707 s.allocCount = nalloc
708 wasempty := s.nextFreeIndex() == s.nelems
709 s.freeindex = 0 // reset allocation index to start of span.
711 getg().m.p.ptr().traceReclaimed += uintptr(nfreed) * s.elemsize
714 // gcmarkBits becomes the allocBits.
715 // get a fresh cleared gcmarkBits in preparation for next GC
716 s.allocBits = s.gcmarkBits
717 s.gcmarkBits = newMarkBits(s.nelems)
719 // Initialize alloc bits cache.
720 s.refillAllocCache(0)
722 // We need to set s.sweepgen = h.sweepgen only when all blocks are swept,
723 // because of the potential for a concurrent free/SetFinalizer.
724 // But we need to set it before we make the span available for allocation
725 // (return it to heap or mcentral), because allocation code assumes that a
726 // span is already swept if available for allocation.
727 if freeToHeap || nfreed == 0 {
728 // The span must be in our exclusive ownership until we update sweepgen,
729 // check for potential races.
730 if state := s.state.get(); state != mSpanInUse || s.sweepgen != sweepgen-1 {
731 print("mspan.sweep: state=", state, " sweepgen=", s.sweepgen, " mheap.sweepgen=", sweepgen, "\n")
732 throw("mspan.sweep: bad span state after sweep")
734 // Serialization point.
735 // At this point the mark bits are cleared and allocation ready
736 // to go so release the span.
737 atomic.Store(&s.sweepgen, sweepgen)
740 if nfreed > 0 && spc.sizeclass() != 0 {
741 c.local_nsmallfree[spc.sizeclass()] += uintptr(nfreed)
742 res = mheap_.central[spc].mcentral.freeSpan(s, preserve, wasempty)
743 // mcentral.freeSpan updates sweepgen
744 } else if freeToHeap {
745 // Free large span to heap
747 // NOTE(rsc,dvyukov): The original implementation of efence
748 // in CL 22060046 used sysFree instead of sysFault, so that
749 // the operating system would eventually give the memory
750 // back to us again, so that an efence program could run
751 // longer without running out of memory. Unfortunately,
752 // calling sysFree here without any kind of adjustment of the
753 // heap data structures means that when the memory does
754 // come back to us, we have the wrong metadata for it, either in
755 // the mspan structures or in the garbage collection bitmap.
756 // Using sysFault here means that the program will run out of
757 // memory fairly quickly in efence mode, but at least it won't
758 // have mysterious crashes due to confused memory reuse.
759 // It should be possible to switch back to sysFree if we also
760 // implement and then call some kind of mheap.deleteSpan.
761 if debug.efence > 0 {
762 s.limit = 0 // prevent mlookup from finding this span
763 sysFault(unsafe.Pointer(s.base()), size)
768 c.local_largefree += size
772 // The span has been swept and is still in-use, so put
773 // it on the swept in-use list.
774 mheap_.sweepSpans[sweepgen/2%2].push(s)
779 // reportZombies reports any marked but free objects in s and throws.
781 // This generally means one of the following:
783 // 1. User code converted a pointer to a uintptr and then back
784 // unsafely, and a GC ran while the uintptr was the only reference to
787 // 2. User code (or a compiler bug) constructed a bad pointer that
788 // points to a free slot, often a past-the-end pointer.
790 // 3. The GC two cycles ago missed a pointer and freed a live object,
791 // but it was still live in the last cycle, so this GC cycle found a
792 // pointer to that object and marked it.
793 func (s *mspan) reportZombies() {
795 print("runtime: marked free object in span ", s, ", elemsize=", s.elemsize, " freeindex=", s.freeindex, " (bad use of unsafe.Pointer? try -d=checkptr)\n")
796 mbits := s.markBitsForBase()
797 abits := s.allocBitsForIndex(0)
798 for i := uintptr(0); i < s.nelems; i++ {
799 addr := s.base() + i*s.elemsize
801 alloc := i < s.freeindex || abits.isMarked()
807 if mbits.isMarked() {
812 zombie := mbits.isMarked() && !alloc
822 hexdumpWords(addr, addr+length, nil)
827 throw("found pointer to free object")
830 // deductSweepCredit deducts sweep credit for allocating a span of
831 // size spanBytes. This must be performed *before* the span is
832 // allocated to ensure the system has enough credit. If necessary, it
833 // performs sweeping to prevent going in to debt. If the caller will
834 // also sweep pages (e.g., for a large allocation), it can pass a
835 // non-zero callerSweepPages to leave that many pages unswept.
837 // deductSweepCredit makes a worst-case assumption that all spanBytes
838 // bytes of the ultimately allocated span will be available for object
841 // deductSweepCredit is the core of the "proportional sweep" system.
842 // It uses statistics gathered by the garbage collector to perform
843 // enough sweeping so that all pages are swept during the concurrent
844 // sweep phase between GC cycles.
846 // mheap_ must NOT be locked.
847 func deductSweepCredit(spanBytes uintptr, callerSweepPages uintptr) {
848 if mheap_.sweepPagesPerByte == 0 {
849 // Proportional sweep is done or disabled.
858 sweptBasis := atomic.Load64(&mheap_.pagesSweptBasis)
860 // Fix debt if necessary.
861 newHeapLive := uintptr(atomic.Load64(&memstats.heap_live)-mheap_.sweepHeapLiveBasis) + spanBytes
862 pagesTarget := int64(mheap_.sweepPagesPerByte*float64(newHeapLive)) - int64(callerSweepPages)
863 for pagesTarget > int64(atomic.Load64(&mheap_.pagesSwept)-sweptBasis) {
864 if sweepone() == ^uintptr(0) {
865 mheap_.sweepPagesPerByte = 0
868 if atomic.Load64(&mheap_.pagesSweptBasis) != sweptBasis {
869 // Sweep pacing changed. Recompute debt.
879 // clobberfree sets the memory content at x to bad content, for debugging
881 func clobberfree(x unsafe.Pointer, size uintptr) {
882 // size (span.elemsize) is always a multiple of 4.
883 for i := uintptr(0); i < size; i += 4 {
884 *(*uint32)(add(x, i)) = 0xdeadbeef